Dev open-sources aletheia-mcp, a deterministic 25µs pre-execution filter blocking prompt-injected MCP tool calls
Fuzzy-Ad9840 · reddit · 2026-09-13
A developer released aletheia-mcp (Apache-2.0, on npm and the MCP registry): a deterministic, no-LLM pre-execution filter that blocks prompt-injection and scope-creep MCP tool calls before they run. It works as standalone guard tools or a fail-closed proxy in front of any MCP server, with 25µs overhead and signatures built from a cross-corpus dataset of real incidents (destructive bash, SQL mutation, filesystem escapes, credential/SSRF exfiltration), plus 155+ tests. The author candidly notes it's a pattern filter, not a sandbox: a real bypass was reported, verified, fixed, and disclosed (GHSA-grqh-xhm8-rj49). They invite discussion on where signature/blacklist approaches hit their ceiling versus allowlists or sandboxed execution.
More from coding & agent
- 'Start more agents' is not an acceptable answer (it is): dev jokes about the idle-time problem — flavioAd · 2026-09-13
- Community ports experimental DeepSeek V4.1 steering into antirez's C-based ds4 CLI — antirez · 2026-09-13
- LangChain's guide to building a custom agent harness: context is everything — iamrobotbear · 2026-09-13
- Nex-N2.5 agentic models hit OpenRouter free tier, built around visual feedback loops — anthara_ai · 2026-09-13
- Dev to stream live demo of hooking an AI agent up to multiple MCPs — PurzBeats · 2026-09-13
- terms.txt paper proposes machine-readable access terms and pricing for AI agents — dair_ai · 2026-09-13